BIBLIOTECA - UAH
Transcript of BIBLIOTECA - UAH
BIBLIOTECA
This work is licensed under a
Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
Document downloaded from the institutional repository of the University of Alcala: http://dspace.uah.es/
This is a postprint version of the following published document:
Guijarro-Díez, M., Nozal, L., Marina, M.L. & Crego, A.L. 2015, "Metabolomic fingerprinting of saffron by LC/MS: novel authenticity markers", Analytical and Bioanalytical Chemistry, vol. 407, no. 23, pp. 7197-7213.
The final publication is available at Springer via: http://dx.doi.org/10.1007/s00216-015-8882-0
© 2015 Springer
(Article begins on next page)
1
Metabolomic fingerprinting of saffron by LC/MS: Novel authenticity markers.
Miguel Guijarro-Díez, Leonor Nozal, María Luisa Marina, Antonio Luis Crego*.
Department of Analytical Chemistry, Physical Chemistry and Chemical Engineering,
University of Alcalá, Ctra. Madrid-Barcelona Km. 33,600, 28871 Alcalá de Henares
(Madrid), Spain.
*Corresponding author
E-mail: [email protected]
Fax: + 34-91-8854971
Keywords: Metabolomic fingerprinting, liquid-chromatography-time of flight-mass
spectrometry, saffron, authenticity markers.
2
Abstract
An untargeted metabolomic approach using liquid chromatography coupled to
electrospray ionization-time of flight-mass spectrometry was developed in this work to
identify novel markers for saffron authenticity which is an important matter related to
consumer protection, quality assurance, active properties, and also economical impact (saffron
is the most expensive spice). Metabolic fingerprinting of authentic and suspicious saffron
samples from different geographical origin was obtained and analysed. Different extracting
protocols and chromatographic methodologies were evaluated to obtain the most adequate
extracting and separation conditions. Using an ethanol/water mixture at pH 9.0 and an elution
gradient with a fused core C18 column enabled obtaining the highest number of significant
components between authentic and adulterated saffron. By using multivariate statistical
analysis, predictive classification models for authenticity and geographical origin were
obtained. Moreover, 84 and 29 significant metabolites were detected as candidates for
markers of authenticity and geographical origin, respectively, from which only 34 metabolites
were tentatively identified as authenticity markers of saffron, but none related to its
geographical origin. Six characteristic compounds of saffron (kaempferol 3-O-glucoside,
kaempferol 3-O-sophoroside, kaempferol 3,7-O-diglucoside, kaempferol 3,7,4´-O-
triglucoside, kaempferol 3-O-sophoroside-7-O-glucoside, and geranyl-O-glucoside) were
confirmed by comparing experimental MS/MS fragmentation patterns with those provided in
scientific literature being proposed as novel markers of authenticity.
3
1. Introduction
Saffron is produced from dried stigmas of Crocus sativus L., and is the most expensive
spice in the world due to the direct labour required for its cultivation, harvesting, and
handling. It has long been used as flavouring and colouring in food preparation, and it is also
known for a wide range of health promoting benefits in traditional and modern medicine [1].
The three main secondary metabolites in saffron are crocins (crocin and its derivatives as
colour factors), a family of yellow pigments freely soluble in water, picrocrocin (taste factor),
a colourless and bitter tasting glycoside and safranal (perfume factor), a volatile oil and the
saffron’s characteristics odour and aroma. In addition, saffron also contains flavonoids,
proteins, sugars, vitamins, amino acids, mineral matter, gums, and other chemical compounds
[2].
Due to its high price and limited production, saffron has been subjected to various
types of adulteration over the centuries. Saffron adulterations are done in order to increase its
weight with foreign matters, and/or to enhance its colour with natural or synthetic colorants to
mask foreign matters addition or to improve their colouring properties. Common fraudulent
practices include the addition of different plant stuffs with similar colour and morphology.
Historically, the most frequently encountered materials have been Crocus sativus stamens
(even styles or strips of the corolla), Carthamus tinctorius L. petals (safflower), Curcuma
longa L. rhizomes (turmeric), Calendula officinalis L. and Arnica montana L. flowers, Bixa
orellana L. seeds, Hemerocallis lilioasphodelus L. petals and Crocus vernus L. stigmas [3, 4].
More recently, a new adulteration by gardenia additions may have reached the European
market. The use of gardenia (extract obtained from the fruits of Gardenia jasminoides J. Ellis)
is a sophisticated method of adulteration, considering that gardenia and saffron differ merely
in the pigments contained [5, 6].
4
Up to the moment, the quality of saffron is certified in the international trade market
following the ISO 3632 normative [7, 8], which certifies by means of a combination of UV
spectrophotometric measurements of picrocrocin at 250 nm and safranal at 310 nm, and
chromatographic profiles of polar dyes and pigments (i.e. crocins) at 440 nm. The measured
parameters allow controlling the quality of saffron through the contents of picrococin,
safranal and crocins as well as the possible presence of some dyes that can be toxic. This ISO
normative is currently under revision to incorporate non-polar dyes and pigments. However,
this update will not solve the important drawbacks of the use of these standards, highlighted
by recent literature, and their weak reliability in the detection of plant foreign matter.
Unfortunately, the ISO 3632 standards are non-specific and unable to separate authentic and
adulterated saffron adequately. In particular, it has been demonstrated that a contamination of
ground saffron with amounts of up to 20% (w/w) of Calendula flowers, safflower or turmeric
was not revealed by the ISO 3632 standards, as it was recently reported [9].
Various authors have proposed several analytical methods for the detection of plant
adulterants in saffron, as UV-Vis spectrophotometric measurements [10-12], near infrared
spectroscopy [13], raman and nuclear magnetic resonance spectroscopy (NMR) [14, 15],
capillary electrophoresis [16], and high-performance liquid chromatography without and with
Mass Spectrometry (MS) detection [17-20]. Most of these methods used to detect adulteration
of saffron are based on targeted analysis of a number of compounds, and their main
disadvantage is that usually only known (targeted) compounds can be detected as indicators
of certain type of adulteration. Finally, the use of molecular methods to detect DNA markers
has been employed so far with encouraging results [21-24]. Low amounts (up to 1%) of plant
adulterants (safflower and turmeric) used as bulking agents were detected as a fraud in
commercial saffron. Nevertheless, there is still an on going demand for the development of
5
faster, simple and robust screening methods suited for identifying saffron adulteration,
especially at levels that make practical economic sense.
In conclusion, despite these attempts for quality control and standardization, there is a
rich history of saffron adulteration constantly evolving. Therefore the search for markers of
authenticity instead of markers of adulteration would be the most intelligent and definitive
strategy to detect adulteration of saffron in which the implementation of metabolomics
approaches provides the tools needed to face this challenge [25]. The general principle of
metabolomics is to characterize biological samples by the production of a chemical signature
or fingerprint. From an analytical point of view, the most widely used technique for this
purpose has been NMR, but MS is becoming more widely used in this field [26, 27]. Indeed,
MS offers higher performance in terms of sensitivity, which is extremely useful for measuring
species with low abundance, as that provides valuable information. Moreover, the specificity
of MS (through high-resolution and/or MS/MS techniques) can help and even facilitate
elucidation of the chemical structures of potential metabolites of interest (i.e. identification of
biomarkers).
These new strategies are also being studied for their application to solve current food
fraud issues where classical methods fail to detect them. Thus, a methodology based on NMR
metabolite fingerprinting has been published very recently proving to be efficient for
determining and identifying fraudulent additions of bulking agents to saffron, considering the
difficulties in detecting saffron fraud according to the ISO 3632 standard methods, especially
when plant adulterants are involved and the spice is commercialized in powder form [28].
Taking into account the deficiency of established methodologies to detect saffron adulteration
with plant adulterants, the method developed could be viable for dealing with saffron frauds
at a minimum level of 20% (w/w) with four typical plant-derived materials employed as
bulking agents in saffron (Crocus sativus stamens, safflower, turmeric and gardenia).
6
The main advantage of metabolomics in food authentication makes use of its
untargeted nature, which can enable the detection of emerging frauds. Thus, the aim of this
work was to explore, for the first time, the feasibility of applying liquid chromatography-
quadrupole time of flight mass spectrometry (LC-(QTOF)MS) together with multivariate
statistical analysis for the assessment of the authenticity of saffron using an untargeted
metabolomics approach. Metabolomics combined with a sensitive analytical technique as MS
and with multivariate analysis is a valid and powerful tool to investigate the quality and
authenticity of saffron. The broad applicability of LC-(QTOF)MS to metabolites of all classes
justified its choice for the problem under consideration. In the present study, authentic saffron
samples from Spain and Iran, as high value samples and the most popular saffron consumed
globally, frequently sold as 100% pure, were chosen as ‘saffron model’ for the study.
2. Materials and Methods
2.1 Chemicals and samples
Acetonitrile, ethanol, methanol, acetic and formic acid of HPLC grade were purchased
from Scharlab (Barcelona, Spain). Sodium borate and phosphoric acid were obtained from
Sigma (St. Louis, MO, USA). Ultrapure water for the chromatographic mobile phase and for
preparing the saffron extracts was obtained from a Milli-Q system (Millipore, Madrid, Spain).
Several standards, verapamil, niflumic acid, propranolol, terfenadine, geranic acid, genistein,
baicalein, quercitrin, rutin, quercetin and kaempferol were purchased from Sigma (St. Louis,
MO, USA).
Ten saffron samples (stigmas or powdered) from Spain and Iran were provided by
“Carmencita" company (Alicante, Spain). Their quality and authenticity had been checked
according to the ISO 3632 parameters and HPLC analysis of dyes. All authentic saffron
7
samples belonged to the commercial category I. In addition, a total of ten saffron samples
suspected of adulteration (stigmas or powdered) purchased in Spain and Iran markets were
also provided by “Carmencita" company. The suspicious samples were considered as such
according to the criteria of the market based on their low cost and/or questionable origin.
2.2 Sample Preparation
Saffron stigmas were finely ground in a mortar with stainless steel balls Ultra Turrax
(IKA, Staufen, Germany) for 2 min. Ground and powdered saffron samples were extracted by
ultrasonic-assisted solid/liquid extraction (0.3 g in 3 ml) using ethanol:borate buffer at pH 9.0
(50:50 v/v) as extractant solvent for 15 min at room temperature. After extraction, samples
were centrifuged for 15 min at 4000 g. 2 mL of supernatant fraction was collected, diluted 1:1
with the extractant solvent, and ultra-filtered through a 3 kDa cut-off filter (Amicon Ultra
Filters, Merck, Darmstadt, Germany) to remove proteins and large molecules.
A quality control (QC) sample was prepared by combining equal aliquots from each
saffron extract (both authentic and suspicious). In the case of untargeted metabolomic
approaches, the performance of the chemometric models could be ensured if a “biological
QC” sample (prepared by combining extracts of all samples employed in the metabolomic
approach) analyzed repeatedly is predicted at the middle of the model [29]. Additionally, a
“test sample” was prepared by adding four standards (verapamil, niflumic acid, propranolol,
and terfenadine) to the QC sample at 0.1 µg/mL in order to characterize the LC-MS system.
2.3 Analytical setup
The analysis was completed using an HPLC system (1100 series, Agilent
Technologies, Waldbronn, Germany) coupled to a quadrupole time-of-flight (QTOF)
equipped with an orthogonal electrospray ionization source (ESI) with Jet Stream thermal
8
focusing technology (6530 series, Agilent) and operating in positive or in negative ion modes
(i.e. polarity-switch was not used, and the samples were analyzed twice). The HPLC system
consisted of a degasser, a quaternary pump, an automatic injector, and a thermostatic column
compartment. Agilent Mass Hunter Qualitative Analysis software (B.04.00) was used for MS
control, data acquisition, and data analysis. A solution of compounds whose masses are
known with great accuracy was continually infused to the system to allow constant mass
correction for accurate mass calibration. Thus, during all analysis, two reference masses were
used, m/z 121.0509 (C5H4N4) and m/z 922.0098 (C18H18O6N3P3F24) for positive ionization
mode, and m/z 112.9856 (C2O2F3(NH4)) and m/z 1033.9881 (C18H18O6N3P3F24) for negative
ionization mode.
2.4 LC-MS conditions
Analysis was completed at 40 ºC using different columns, all of them Ascentis
Express (Sigma, St. Louis, USA), with the same dimension (100 x 2.1 mm i.d.) and type of
packed bed (fused-core® particles with 0.5 µm thick porous shell and an overall particle size
of 2.7 µm) but of different kinds of stationary phases (C18, Cyano or HILIC). In addition,
Ascentis Express guard columns (5 x 2.1 mm i.d.) of the same material as the analytical
column in each case were employed. The system was operated with an injected volume of 15
μL and a flow rate of 0.4 mL min-1. The mobile phases consisted of 0.1% formic acid for
ESI+ or 10 mM ammonium formate for ESI- in Milli-Q water (eluent A) and 0.1% formic
acid for ESI+ or 10 mM ammonium formate for ESI- in acetonitrile (eluent B), using different
elution gradients according to the selected stationary phase. For the C18 and Cyano columns,
the linear gradient started from 5% B to 95% B in 33 min and returned to starting conditions
in 1 min, keeping the re-equilibration at 5% B for 15 min. For the HILIC column the same
linear gradient profile was used but eluent A was 0.1% formic acid for ESI+ or 10 mM
9
ammonium formate for ESI- in acetonitrile and B was 0.1% formic acid for ESI+ or 10 mM
ammonium formate for ESI- in Milli-Q water.
The ionization source conditions were as follows: capillary voltage of 3 kV with a
nozzle voltage of 0 V; nebulizer pressure at 2.7 bar; sheath gas of jet stream of 6.5 L/min at
300 ºC; and drying gas of 10 L/min at 300 ºC. The cone voltage (fragmentator) after sampling
capillary was set at 175 V. The skimmer and octapole voltages were 60 V at 750 V,
respectively.
MS analysis was performed in both positive and negative ESI modes with the mass
range set at m/z 100-1700 (extended dynamic range) in full scan resolution mode at a scan rate
of 2 scans per second. These conditions allowed to reach an average mass resolution of
0.0001 Da calculated by full width half maximum of standards in the “test sample”. Also, to
characterize the LC-MS system in terms of mass accuracy and reproducibility, the "test
sample" was injected at the beginning, middle, and end of the analysis sequence.
For sample analysis, replicate extractions (n = 3) of all samples were used in random
sequence to ensure that any experimental trends observed were directly associated with the
sample and not due to any change in the instruments performance or sample preparation over
time. Also, QC sample was injected at the beginning of the run and after every 3 real samples
analysis to ensure the stability and repeatability of the LC–MS system.
2.5 Data treatment
The use of dedicated software solutions for handling the large datasets typically
produced in metabolomics is unavoidable. Hence a data processing to convert the initial
three-dimensional raw data (m/z, retention time, and intensity of ion current) to a two-
dimensional data table reporting time-aligned and mass-aligned abundances of
chromatographic peaks was performed. First, the resulting data file was cleaned of
10
background noises and unrelated ions by the Molecular Feature Extraction (MFE) tool in
Mass Hunter Qualitative Analysis software. MFE algorithm uses the accuracy of the mass
measurements to group ions related by charge-state envelope, isotopic distribution, and/or the
presence of adducts and dimmers. The MFE then creates a listing of all possible components
(molecular features) as represented by the full TOF mass spectral data. For data extraction by
MFE, the molecular features present in a sample were determined using the following
parameters in the software: target data type of small molecules (chromatographic); peaks with
height ≥ 500 counts; peak spacing tolerance = 0.0025 m/z, plus 7.0 ppm; isotope model =
common organic molecular; limited assigned change = 3; and then, to find co-eluting adducts
of the same feature the following adduct settings were applied: H+, Na+, K+, and NH4+ in
positive ionization, and HCOO- for negative ionization. Second, other data processing such as
filtering and alignment were performed with Mass Profiler Professional software (B.02.00,
Agilent). The retention time (RT) and m/z alignment of the respective peaks was carried out
using the following parameters depending on the mass accuracy and reproducibility obtained
in the analysis of the “test sample”: initial retention time 3 min, final retention time 33 min,
mass tolerance 0.02 Da, mass window 0.02 Da, and retention time window 0.1 min. To clean
the data matrix from random signals and to select only features with biological meaning filter
by frequency was applied. Choosing the data present in all quality control samples performed
primary filtering. Also, features were filtered by choosing masses that were present in all
samples at least in one of two groups for comparison authentic vs. suspicious and Spanish vs.
Iranian. A secondary filtering was performed by choosing the data with a coefficient of
variation below 35%.
Further data pretreatment was performed to the data sets obtained from the previous
data-processing stage before application of statistics. First, the data sets were transformed by
applying common logarithm to intensities in order to reduce the influence of a few
11
particularly intense signals that can strongly influence statistical analysis and subsequent
interpretation. This aspect is of particular concern in metabolomics, due to the huge dynamic
range in terms of metabolite-concentration levels. Second, the data sets were pre-treated using
the Pareto scaling (the square root of the standard deviation is used as the scaling factor).
The data tables generated in the previous stage of data processing was analyzed
comprehensively with appropriate statistical tools. Thus, data analysis was performed with
univariate methods using Microsoft Excel software (2010, Redmon, WA, USA) to reveal
potential candidate compounds with significant differences in terms of abundance between
two groups of samples (authentic and suspicious). The normality of distribution for each data
set was assessed using the “Wilk-Shapiro’s test”, the homogeneity of variances was studied
using the “Levene´s test”, and “t-test” (assumed equal variance) or “Welch’s test” (assumed
unequal variance) could reveal potential candidate compounds with significant differences in
terms of abundance between two groups of samples (authentic and suspicious). Likewise,
multivariate statistical analysis was performed on mass spectral data sets using SIMCA-
P+12.0 software (Umetrics, Umeå, Sweden). Unsupervised principal component analysis
(PCA) was applied to represent the sample distribution in the multivariate space. Supervised
partial least squares-discriminant analysis (PLS-DA) and orthogonal partial least squares-
discriminant analysis (OPLS-DA) were used in order to reduce the model complexity by
removing the systematic variations in the X matrix that were not related to Y response
(structured noise) maximizing the separation among samples.
The quality of the models was described by the goodness of fit (R2 value), and the
predictive ability (Q2 value). Recognition ability (R2) represents a percentage of successfully
classified samples in the training set. Prediction ability (Q2) is a percentage of correctly
classified samples in the test set by using the model developed during the training step. In
addition, to ensure the performance of the models, QC analysed at the beginning of the run
12
and after every 3 real samples analysis should be predicted at the middle of the model since it
was prepared by combining equal aliquots from each saffron extract (both authentic and
suspicious). Also, to avoid the risk of over fitting for a discriminant analysis models used for
the selection of statistically significant metabolites according to jack-knifed confidence
intervals, models were validated by the use of a cross-validation tool using a 1/3 out approach
[29, 30]. The data set was divided into three parts and 1/3rd of samples were excluded to build
a model with the remaining 2/3rd of samples. This new model then predicted excluded
samples and the procedure was repeated until all samples had been predicted at least once.
The percentage of correctly classified samples was calculated each time.
Finally, selection of potential biomarkers was maintained for each comparison based
on different tests: (i) “t-test” or “Welch’s test” depending on the homogeneity of variances,
calculated by using Microsoft Excel; and (ii) S-plot and jack-knife confidence intervals
obtained for OPLS-DA models in SIMCA-P+12.0.
2.6 Databases for Identification
Metabolites were tentatively identified by searching by mass accuracy against the
online available databases such as the METLIN (http://metlin.scripps.edu), HMDB
(http://hmdb.ca), KEGG (http://genome.jp/kegg), FooDB (http://foodb.ca/), and lipidMAPS
(http://www.lipidmaps.org/data/databases.html). For features, individual searching in
databases was performed employing an error of 10 ppm.
3. Results and discussion
3.1 LC-MS metabolomic analysis
The extremely wide diversity of potential metabolites present in a sample in terms of
chemical structures and concentrations means that it is unrealistic to have the goal of
13
measuring all of them in metabolomics. Thus, minimum sample preparation is preferred,
especially for untargeted applications without any presupposed hypothesis. For solid matrices
such saffron samples an extraction step is required for transferring the metabolome
compounds into a liquid phase [31], and a subsequent protein-elimination step to limit ion
suppression when using ESI for liquid chromatography with MS is employed [32].
Considering the fact that any sample treatment step can potentially result in its alteration and
consecutive losses of some metabolites, only three sample preparation steps after extraction,
centrifugation, dilution and ultrafiltration, essential to avoid instrument problems like column
clogging or MS system contamination, were carried out.
In order to obtain the greatest number of compounds from saffron samples, the nature
of extraction solvent was considered during the extraction procedure. As the saffron
compounds can have varied polarities, different extraction solvents were investigated using
different mixtures between ethanol (0, 50 and 100 %) and aqueous buffer at two different pHs
(low pH at 2.5 and high pH at 9.0). Also, chromatographic separation was investigated using
different stationary phases, two reversed phases (C18 and CN) and one polar phase (HILIC),
to allow separation of a range of compounds of different polarities, from low to high polarity.
In addition, generic HPLC methods (see section 2.4) were applied to cover a wide range of
metabolites with diverse chemical and physical properties. Finally, MS analysis was done in
both positive and negative ionization modes to ensure that metabolites extracted from saffron
samples amenable to either positive or negative ionization were covered, in order to monitor
as many ions as possible.
To evaluate the different analytical approaches, the total number of possible
components (molecular features) present in an authentic and other suspicious saffron sample
was compared when the different conditions described before were used in the MS analysis
for both positive and negative ionization modes (see Table 1). Initially, ethanol and aqueous
14
buffer at different pH (2.5 and 9.0) were investigated as extraction solvents to increase the
efficiency of saffron components extraction. As shown in Table 1, ethanol and borate buffer
at pH 9.0 worked better, so the complementarity of both was tested. A substantial increase in
the number of features was obtained with the mixture ethanol and borate buffer at pH 9.0
(50:50 v/v). On the other hand, the information obtained using the reversed phase columns
(chromatographic separation and features) was more useful than data obtained using the
HILIC column. Between the reversed phase columns, the C18 was the best option due mainly
to better resolution between features, since the number of molecular features extracted was
only increased by 110%, being this number between 30-60% less in ESI- than ESI+ (see
Table 1). In conclusion, the mixture ethanol and borate buffer at pH 9.0 (50:50 v/v) as
extraction solvent and a C18 column were chosen for further studies, using MS analysis in
both positive and negative ion modes to expand the possibilities of discovering new markers.
On the other hand, the influence of extraction time in sonic bath was also investigated.
Samples were extracted for 5, 10, 15 and 20 min. Similar results were obtained with the
extraction time of 15 and 20 min, therefore 15 min was chosen as the optimum extraction
time (results not shown).
Finally, other ESI parameters were also studied: i) only depending on analytes as
nozzle voltage (0-1000 V) and fragmentator voltage (100-200 V), and ii) depending on
mobile phase flow-rate and composition but also limited by analytes thermal stability: drying
gas temperature (200-350 ºC) and sheath gas temperature (250-400 ºC). The optimized ESI
parameters (results not shown) obtained with the previously selected mobile-phase
composition and a flow-rate of 0.4 mL/min were: nozzle voltage, 0 V; fragmentator voltage
175 V; drying gas flow-rate, 10 L/min; drying gas temperature, 350 ºC; sheath gas flow-rate,
7.5 L/min, and sheath gas temperature, 350 ºC.
15
Table 1. Total number of possible components (molecular features) present in an authentic and other suspicious saffron samples obtained by MFE software when different extraction solvents for sample preparation and different chromatographic columns were used in the MS
analysis for both positive and negative ionization modes. Experimental conditions as in section 2.4.
C18
Phosphoric buffer
pH2.5
Borate buffer
pH 9.0 Ethanol
Ethanol/borate buffer pH 9.0
(50:50 v/v)
SAMPLES ESI+ ESI- ESI+ ESI- ESI+ ESI- ESI+ ESI-
Authentic 9663 6764 9975 6987 10056 6164 14673 7588 Suspicious 8438 4586 9380 3996 1223 648 10304 4578
CN
Phosphoric buffer
pH2.5 Borate buffer
pH 9.0 Ethanol
Ethanol/borate buffer pH 9.0 (50:50 v/v)
SAMPLES ESI+ ESI- ESI+ ESI- ESI+ ESI- ESI+ ESI-
Authentic 8857 5677 9344 6254 9754 5632 12456 7040
Suspicious 8336 4245 9055 3327 1275 543 9687 4176
HILIC
Phosphoric buffer
pH2.5 Borate buffer pH
9.0 Ethanol
Ethanol:borate buffer pH 9.0 (50:50 v/v)
SAMPLES ESI+ ESI- ESI+ ESI- ESI+ ESI- ESI+ ESI-
Authentic 3452 1085 3654 1132 3987 1129 4967 1235 Suspicious 2968 354 3088 376 1234 166 4674 476
3.2 Chemometric analysis
Chromatograms from all saffron samples and QCs using positive and negative
ionization modes were aligned, revealing a number of features in total of 25645 and 17153,
respectively. The PCA represents a highly useful tool when interpreting complex multivariate
data sets, as it allows dimensionality reduction and visualization of the information present in
the original data in the form of a few principal components while retaining the maximum
possible variability [33]. As shown in Figure 1, score plots from unsupervised PCA of the
data set representing all saffron samples and QC samples showed more pronounced clustering
and significantly better differentiation among sample clusters for positive ionization data (see
Figure 1A) compared to those acquired in the negative mode (see Figure 1B). Therefore,
only data acquired in positive ionization mode were further used in this statistical evaluation
with univariate and multivariate methods, as it was found fit for purpose, in addition as
16
discussed above, the ESI+ gave almost a 200% increase in molecular features compared to the
ESI-.
(A)
(B)
Figure 1. PCA score plot obtained in positive (A) and negative (B) ionization mode data from saffron
metabolic profiles obtained for authentic samples, samples suspected of adulteration, and QC samples.
Suspicious samples. Authentic samples. ● QC samples.
For quality checking, a PLS-DA model of positive ionization data was built for the
two groups of authentic and suspicious samples, taking into account all variables generated
17
from the mass spectra. As it can be observed in Figure 2, even without any filtering or
scaling, the samples were clustered clearly, and the quality of the model built for two
components was very good with excellent values for variance explained (R2X = 0.990, and
R2Y = 0.986) and variance predicted (Q2 = 0.975). The robustness of the analytical procedure
was tested by prediction of the QC samples in the model, and proved by the tight clustering of
the QC samples in the middle of the plot (see Figure 2). Since QC samples were obtained by
mixing equal volumes of all the samples, the model proved that separation between groups
was not random but due to real variability. In addition, the fact that QCs clustered together
proved the stability and repeatability of the methodology, confirming that clustering (or
separation) was due to the sample content and not to the analytical conditions and therefore,
data are suitable for further statistical analysis.
Figure 2. Score plot for a PLS-DA model built with the whole data of saffron metabolic profiles
obtained for authentic samples and samples suspected of adulteration with prediction for QC samples.
Suspicious samples. Authentic samples. ● QC samples.
OPLS-DA model, which comes from PLS-DA model, was employed to discriminate
between groups of samples and to obtain the significance of the discriminatory compounds by
18
using jack-knife test. In OPLS-DA a regression model is calculated between the multivariate
data and a response variable that only contains class information. The advantage of OPLS-DA
compared to PLS-DA is that a single component is used as a predictor for the class, while the
other components describe the variation orthogonal to the first predictive component [34-36].
To perform the OPLS-DA model for saffron samples classification, chromatograms from
every saffron sample (authentic samples and samples suspected of adulteration) were aligned
with the same parameters described in the Material and Methods section. First, features were
filtered by choosing the data present in 100 % of QC samples and present in at least one of
two groups for comparing authentic vs. suspicious. 1126 features (out of 24768) were selected
after primary filtering and chosen for further filtering. In total, 490 different features fulfilled
the requirements of a secondary filtering performed by choosing the data with a coefficient of
variation below 35%.
The OPLS-DA model with two groups (authentic and suspicious) was built with those
490 features that fulfilled the filtering requirements and using pareto scaling [37]. As shown
in Figure 3A, the quality of the model built was excellent regarding variance explained (R2X
= 0.999, and R2Y = 0.991) and variance predicted (Q2 = 0.984). In addition, the large
orthogonal variation divided the authentic samples into two groups, both clearly differentiated
from suspicious samples. Reviewing the origin of authentic samples, it was observed that
each group corresponded to Spanish and Iranian saffron. Therefore, new OPLS-DA model
was performed according to the geographical origin of the authentic samples (see Figure 3B).
In that case, suspicious samples were excluded from the model. The same parameters for
alignment and filtering were employed to perform the OPLS-DA model for origin saffron
classification. In this case, 1127 features (out of 19351) were selected after primary filtering
and chosen for further filtering. In total, 489 different features fulfilled the requirements of a
19
secondary filtering performed. Again, the quality of the model built was excellent regarding
variance explained (R2X = 0.998, and R2
Y = 0.989) and variance predicted (Q2 = 0.982).
(A)
(B)
Figure 3. OPLS-DA score plot of saffron metabolic profiles obtained for: (A) authentic samples and samples
suspected of adulteration, and (B) authentic saffron samples from Spain and Iran.
Suspicious samples. Authentic samples. Spanish authentic samples. + Iranian authentic samples.
20
In conclusion, separations by authenticity (authentic and suspicious) and by origin
(Spain and Iran) of saffron samples were possible. In addition, to avoid the risk of over fitting,
both models (authenticity and origin) were validated by cross-validation tool as described in
the Materials and Methods section, and the percentages of samples correctly classified for
models (authenticity and origin) were tested [38]. In the validation of authenticity model, all
the authentic samples were 100 % predicted with appropriate classification, and from
suspicious samples only one out of 10 samples was misclassified. In the validation of origin
model, all the Iranian samples were 100 % predicted with appropriate classification, and from
Spanish samples only one out of 5 samples was misclassified enabling these results to be
considered satisfactory.
Finally, all potential unique and high intensity markers giving a significant
contribution were selected with unpaired “Welch’s test” performed for log-transformed
intensities and statistical significance by s-plot and jack-knifed confidence intervals of both
models. When the interval includes the zero value, the covariance is not significant and the
compound should not be considered as a potential marker (see Figure 4). For authentic vs.
suspicious (see Figure 4A) and spanish vs. iranian (see Figure 4B) comparisons, a total of 84
and 29 metabolites (with p-value < 0.05) were obtained as candidates for markers of
authenticity and origin, respectively. It is noteworthy for authentic vs. suspicious comparison
that all significant markers obtained by OPLS-DA model were up regulated in authentic
group according to the statistical significance of these metabolites using the jack-knife
interval (see Figure 4A). This fact implies that the suspicious samples were not adulterated
with common compounds that may serve for discriminant analysis by OPLD-DA and
therefore the significant markers obtained were compounds with higher content in authentic
saffron, i.e, they are markers of authenticity. However, for the spanish vs. iranian
21
Co
vari
ance
AUTHENTIC SUSPICIOUS
Co
vari
ance
SPANISH IRANIAN
comparison, the significant markers obtained by OPLS-DA model were regulated in both
groups according to the statistical significance of these metabolites using the jack-knife
interval (see Figure 4B), that is, there are both significant markers for Spanish samples and
for Iranian samples.
(A)
(B)
Figure 4. Covariance for discriminant variables including jack-knife interval (p > 95%) in (A) authentic samples
and samples suspected of adulteration, and (B) authentic saffron samples from Spain and Iran.
22
3.3 Identification of marker compounds
The identification of discriminating marker compounds represents probably the most
laborious and time-consuming step of the metabolomic workflow. The accurate mass
information was used to propose the elemental formula of each marker using a database
search (see 2.6 section). The search for the elemental formula can be complicated because
multiple hits can be matched with the same exact mass. For reliable elemental formula
estimation, the suggested formulas are filtered based on matching of experimental and
theoretical isotopic profile in terms of relative intensities. Finally, out of the numerous
compound hits obtained, only those whose presence was probable in plants were taken into
account (e.g., hits of drugs and synthetic compounds obtained in databases were cleared).
Using this procedure, 34 of 113 marker compounds (from both models, authenticity and
origin) were tentatively identified. It should be highlighted that no marker related to origin
model was tentatively identified. Therefore, the 34 possible metabolites tentatively identified
were all related to the authenticity model, i.e, they are authenticity markers of saffron.
Table 2 reports all significant metabolites tentatively identified detected as candidates
for authenticity markers of saffron and includes their retention time, the m/z value obtained in
the LC-TOF system and the type of ion responsible for it, the calculated mass error when
comparing with the database, the probable ion elemental formula, the percentage change
among the groups and its statistical significance (p-value), and their coefficient of variance
according to QC. All suggested formulas have very good mass accuracy, usually less than 2
mDa, thus increasing the confidence in predicted formulas.
It should be noted that a mass match with a metabolite in the database does not
provide a conclusive identification. The confirmatory analysis of a predicted compound
standard is generally required for full and unequivocal identity confirmation by comparison of
23
Table 2. Overview of significant metabolites (tentatively identified) detected as candidates to be authenticity markers in saffron.
Tentative identification RT
(min)
Experimental
m/z Ion
Mass error
(ppm)
Mass error
(mDa)
Probable ion
elemental formula
Fold change
(p-value)
CV for QCs
[%]
Kaempferol 3,7,4'-triglucoside 4.8 773.2134 [M+H]+ -0.1 -0.1 C33H41O21 37 (0.007) 9
Kaempferol 3-sophorotrioside [M+H]+ C33H41O21
Kaempferol 3-sophoroside-7-glucoside [M+H]+ C33H41O21
Kaempferol 3-glucoside-7-sophoroside [M+H]+ C33H41O21
Cyanidin 3,5,3'-triglucoside M+ C33H41O21
Delphinidin 3-rutinoside-5-glucoside M+ C33H41O21
Nepetalic acid 4.9 185.1176 [M+H]+ 5.9 0.4 C10H17O3 6 (0.005) 7
Kaempferol 3,7,4'-triglucoside 5.9 773.2142 [M+H]+ 1.0 0.8 C33H41O21 24 (0.01) 12
Kaempferol 3-sophorotrioside [M+H]+ C33H41O21
Kaempferol 3-sophoroside-7-glucoside [M+H]+ C33H41O21
Kaempferol 3-glucoside-7-sophoroside [M+H]+ C33H41O21
Cyanidin 3,5,3'-triglucoside M+ C33H41O21
Delphinidin 3-rutinoside-5-glucoside M+ C33H41O21
Kaempferol 3,7-diglucoside 6.3 611.1601 [M+H]+ -0.9 -0.6 C27H31O16 27 (0.001) 6
Kaempferol 3-sophoroside [M+H]+ C27H31O16
Quercetin 3-rutinoside (Rutin) [M+H]+ C27H31O16
Cyanidin 3,5-diglucoside (Cyanin) M+ C27H31O16
Cyanidin 3-sophoroside M+ C27H31O16
Geranic acid 6.4 169.1448 [M+H]+ 5.3 0.9 C10H17O2 21 (0.002) 5
Dihydrojasmonic acid, Methyl ester 7.5 227.1636 [M+H]+ -1.3 -0.3 C13H23O3 13 (0.009) 10
24
Table 2.Continued.
Tentative identification RT
(min)
Experimental
m/z Ion
Mass error
(ppm)
Mass error
(mDa)
Probable ion
elemental formula
Fold change
(p-value)
CV for QCs
[%]
Kaempferol 3,7-diglucoside 7.7 611.1604 [M+H]+ -0.4 -0.3 C27H31O16 16 (0.003) 8
Kaempferol 3-sophoroside [M+H]+ C27H31O16
Quercetin 3-rutinoside (Rutin) [M+H]+ C27H31O16
Cyanidin 3,5-diglucoside (Cyanin) M+ C27H31O16
Cyanidin 3-sophoroside M+ C27H31O16
Kaempferol 3-glucoside 8.6 449.1050 [M+H]+ -2.0 -0.9 C21H21O11 29 (0.004) 4
Quercetin 3-rhamnoside (Quercitrin) [M+H]+ C21H21O11
Angoluvarin 9.8 485.1991 [M+H]+ 2.6 1.3 C30H29O6 4 (0.009) 15
Isococculidine 9.9 286.1792 [M+H]+ 8.1 2.3 C18H24NO2 11 (0.005) 13
Isobrucein A 9.9 523.2172 [M+H]+ -0.4 -0.2 C26H35O11 4 (0.007) 9
Apimaysin 9.9 561.1618 [M+H]+ 2.1 1.2 C27H29O13 9 (0.008) 11
4,2-Dihydroxy-4,6-
dimethoxychalcone 4-apiosyl-
glucoside
10.1 595.2038 [M+H]+ 2.5 1.5 C28H35O14 9 (0.03) 20
Eriojaposide B 10.4 517.2631 [M+H]+ -3.5 -1.8 C25H41O11 9 (0.04) 18
Cinncassiol-glucoside 10.4 531.2810 [M+H]+ 1.9 1.0 C26H43O11 9 (0.006) 16
Kaempferide 3,7-dirhamnoside 10.6 593.1859 [M+H]+ -1.0 -0.6 C28H33O14 12 (0.009) 9
5-Hydroxypseudobaptigenin 7-O-
Glucoside 11.3 461.1085 [M+H]+ 1.3 0.6 C22H21O11 7 (0.007) 17
Karatavigenin B 11.4 569.3478 [M+H]+ 0.9 0.5 C34H49O7 15 (0.003) 18
Anhalonidine 11.6 224.1285 [M+H]+ 1.8 0.4 C12H18NO3 11 (0.007) 12
Nupharamine 11.6 252.1231 [M+H]+ 0.0 0.0 C13H18NO4 7 (0.04) 15
25
Tentative identification RT
(min)
Experimental
m/z Ion
Mass error
(ppm)
Mass error
(mDa)
Probable ion
elemental formula
Fold change
(p-value)
CV for QCs
[%]
Resokaempferol 11.8 271.0605 [M+H]+ 1.5 0.4 C15H11O5 21 (0.007) 9
Baicalein [M+H]+ C15H11O5
Genistein [M+H]+ C15H11O5
Dihydrojasmonic acid 12.4 213.1491 [M+H]+ 2.3 0.5 C12H21O3 8 (0.004) 10
Aconine 12.8 500.2834 [M+H]+ 4.2 2.1 C25H42NO9 10 (0.009) 8
3-Hydroxyethylbacteriochlorophyllide A 12.9 635.2700 [M+H]+ 1.6 1.0 C35H39MgN4O6 6 (0.04) 13
4,6,8-Megastigmatriene 13.7 177.1645 [M+H]+ 3.9 0.7 C13H21 3 (0.0007) 6
1-O-beta-D-Glucopyranose 13.9 361.1984 [M+H]+ -7.5 -2.7 C17H29O5 26 (0.005) 10
28-Hydroxyglycyrrhetic acid 14.4 487.3394 [M+H]+ -4.9 -2.4 C30H47O5 9 (0.007) 13
Cinnamylisovalerate 14.5 219.1377 [M+H]+ -1.4 -0.3 C14H19O2 7 (0.004) 9
Fissinolide 14.5 513.2438 [M+H]+ 8.0 4.1 C29H37O8 14 (0.03) 12
15,16-Dihydrobiliverdin 14.5 585.2666 [M+H]+ -4.7 -2.8 C32H41O10 6 (0.007) 10
Octadecanedioic acid 18.4 315.2504 [M+H]+ 7.3 2.3 C18H35O4 6 (0.007) 11
Hexafluoro-25-hydroxycholecalciferol 26.5 509.2824 [M+H]+ -4.9 -2.5 C27H39F6O2 7 (0.004) 9
1-octadecatrienoyl-2-
octadecatetraenosyl-glycero-3-phosphate 29.5 691.4350 [M+H]+ 2.5 1.7 C39H63O8P 11 (0.01) 9
4-dimethylaminophenyl-25
dihydroxycholecalciferol 30.9 536.4104 [M+H]+ 0.9 0.5 C35H53NO3 3 (0.04) 15
26
retention time, accurate mass data and isotopic profile of commercially available reagents
with those obtained in real samples. However, metabolites frequently have very complex
chemical structures and may be difficult to synthetize, so commercial standards are not
available or have huge prices.
Many of the compounds grouped in Table 2 are flavonols and anthocyanins
(molecules responsible of color) and substances responsible of flavor. This fact is in
concordance with the main expected differences in a saffron sample adulterated, since the
objective of adulteration is to mask saffron properties such as color and flavor. However, in
several cases different marker compounds were matched with the same elemental formula.
Note especially those having a retention time of 4.8, 5.9, 6.3, 7.7, and 8.6 min, corresponding
to compounds naturally conjugated with sugars (glycosides) with three, two and one hexoses,
respectively. Due to the lack of supplier or huge prices of standards of the majority of these
glycosides, they were confirmed by studies focused on MS and MS/MS analysis in the
positive and negative modes (see Table 3), and comparing experimental MS/MS
fragmentation patterns with those provided in scientific literature.
According to Table 3, the tentatively identified compound as delphinidin 3-rutinoside-
5-glucoside that could elute at retention time 4.8 or 5.9 min was discarded due to the absence
of an ion in positive mode with m/z 303.0499 which should appear as M+ for its aglycone
delphinidin. Likewise, the quercetin glycosides (rutin at 6.3 or 7.7 min, and quercitrin at 8.6
min) were discarded because in all cases signals with m/z about 287 (287.0532, 287.0537 or
287.0527) were obtained in positive mode instead of an ion with m/z 303.0499 corresponding
to its protonated aglycone quercetin. In addition, the standards of the rutin and quercitrin were
analyzed for further confirmation of their absence by their retention times and MS/MS
fragmentation data.
27
Table 3. MS and MS/MS data (in positive and negative ESI modes) for several compounds tentatively identified in Table 2 as glycosides.
RT
(min)
ESI + ESI -
MS ions m/z (%)a MS/MSb ions m/z (%) MS ions m/z (%) MS/MS ions m/z (%)
4.8 773.2076 (...) [M+H]+ 611.1546 (...) [M-162+H]+
611.1543 (...) [M-162+H]+ 449.1049 (...) [M-324+H]+ 287.0531 (...) [M-486+H]+
771.1811 (...) [M-H]- 609.1387 (...) [M-162-H]-
609.1386 (...) [M-162-H]- 447.0854 (...) [M-324-H]- 284.0263 (...) [M-487-H]-
5.9 773.2081 (...) [M+H]+
611.1541 (...) [M-162+H]+
611.1539 (...) [M-162+H]+ 449.1051 (...) [M-324+H]+ 287.0529 (...) [M-486+H]+
771.1880 (...) [M-H]- 609.1374 (...) [M-162-H]-
609.1375 (...) [M-162-H]- 429.0736 (...) [M-342-H]- 284.0258 (...) [M-487-H]-
6.3 611.1543 (...) [M+H]+ 449.1056 (...) [M-162+H]+ 287.0532 (...) [M-324+H]+
609.1371 (...) [M-H]-
447.0824 (...) [M-162-H]- 285.0331 (...) [M-324-H]- 284.0253 (...) [M-325-H]-
7.7 611.1538 (...) [M+H]+ 449.1054 (...) [M-162+H]+ 287.0537 (...) [M-324+H]+
609.1345 (...) [M-H]-
429.0729 (...) [M-180-H]- 284.0259 (...) [M-325-H]-
8.6 449.1057 (...) [M+H]+ 287.0527 (...) [M-162+H]+ 447.0752 (...) [M-H]- 284.0250 (...) [M-163-H]-
a Percentage of signal intensity (in parentheses)
b MS/MS experiments using as precursor ion the quasimolecular ion of MS experiments ([M+H]+ or [M-H]-)
28
The other anthocyanins tentatively identified that eluted at retention times 4.8, 5.9, 6.3 and 7.7
min were also discarded because although signals with m/z about 287 (287.0531, 287.0529,
287.0532 or 287.0537) appeared in the positive mode as M+ for their aglycone cyanidin, the
ion at m/z 286.0483 as [M-H]- was absent in the negative mode. Therefore, the ions at m/z
about 287 in positive mode corresponded to a protonated aglycone of kaempferol. These
results confirmed the well-known fact that kaempferol glycosides represent 84.0% of total
flavonol content in flowers of Crocus sativus L. whereas quercetin and isorhamnetin
glycosides represent only 9.3 and 8.7%, respectively [39, 40].
The tentative identification of the different glycosides of kaempferol listed as
significant authenticity markers of saffron in Table 2 was confirmed on the basis of MS/MS
fragmentation patterns by comparison with previously published data. These studies were
focused on MS analysis in the negative mode (see Table 3) because this mode is more
sensitive than positive mode to establish the differences between interglucosidic linkages (see
MS/MS fragments in Table 3 for compounds that eluted at 4.8 and 5.9 min, and for
compounds that eluted at 6.3 and 7.7 min), and positional isomers of O-glycosylated
kaempferols [41]. Thus, the identification of these markers was achieved in a similar manner,
and therefore, the discussion was focused on two characteristic examples of the five possible
markers, the kaempferols glycosylated with two hexose residues detected at 6.3 min and 7.7
min.
To differentiate kaempferols with the same degree of glycosylation, the
characterization of the (glucosyl(1 → 2)glucosides) interglycosidic linkage was defined by
the presence at 7.7 min of the fragment ion [M-180-H]- characteristic of sophoroside
flavonoids (at m/z 429.0729 formed from the loss of the terminal sugar) although at low
abundance (see Figure 5A), which is absent in a gentiobioside flavonoid (glucosyl (1 → 6)
glucosides) and in a kaempferol di-O-glucoside with two sugar moieties linked to different
29
phenolic hydroxyl positions of the kaempferol nucleus. In addition, the MS fragmentation
pattern of kaempferol 3-O-diglycosides was detected by the presence of the ion [M-162-163-
H]- (m/z 284.0259) as base peak, ion obtained as a result of the loss of two sugar moieties.
The presence of a fragment ion at m/z 284.0259 corresponding to a kaempferol moiety,
instead of the expected ion at m/z 285 (loss of two glucosyl units, [M-162-162-H]-), was
recently described as a characteristic fragment ion for the kaempferol 3-O-glucoside ([M-163-
H]-), but not in the case of kaempferol 7-O-glucoside [43 42]. All these data (see Table 3 and
Figure 5A), according to previously published data [41], confirmed the structure of
kaempferol 3-O-sophoroside as the compound eluting at 7.7 min, the main saffron flavonoid
(about 55% of total flavonols content), whose presence was reported previously in Crocus
sativus in several works [40, 42-46]. In addition, Carmona et al [47] stated that saffron
samples from different geographical origin were clearly separated by their kaempferol 3-
sophoroside contents. However, according to the results obtained in this work, kaempferol 3-
O-sophoroside was not detected as a significant marker of the origin of saffron because its
interval of covariance includes zero value according to the jack-knife interval.
On the other hand, the MS/MS spectrum of the [M-H]- ion (m/z 609.1317) provided
the ion [M-162-H]- (m/z 447.0824) as base peak (see Figure 5B), which was formed by a loss
of one glucose, and the ion [M-162-162-H]- (m/z 285.0331) formed by the loss of two
glucoses. However, the characteristic ion [M-180-H]- of the sophoroside of kaempferol was
not detected. These results confirmed that it is a kaempferol di-O-glucoside with sugar
moieties linked to different phenolic hydroxyl positions of the kaempferol nucleus. This sugar
substitution takes place at the hydroxyls in the following order of preference (from more to
less) at the 3-, 7- and 4'- positions of the flavonoid nucleus [48]. Therefore, these data,
according to previously published data [41], confirmed the structure of kaempferol 3,7-O-
30
diglucoside as the compound eluting at 6.3 min. The presence of this compound was also
described in Crocus sativus in two recent papers [40, 42].
Figure 5. MS/MS spectra in ESI- for compounds eluted at 7.7 min (A) and at 6.3 min (B)
using as precursor ion the deprotonated molecular ion of the kaempferols glycosylated with
two hexoses, at m/z 609.1345 in (A) and at m/z 609.1317 in (B).
With respect to other glycosylated kaempferols tentatively identified, compounds eluted
at retention times of 4.8, 5.9 and 8.6 min, their MS and MS/MS data are also listed in
Table 3.
31
For the last of them (at 8.6 min), its MS spectrum showed the deprotonated molecular ion at
m/z 447.0752, characteristic of a kaempferol glycosylated with only one hexose. This fact
together with the presence of a MS/MS fragment ion at m/z 284.0250 confirmed that it was
kaempferol 3-O-glucoside, according to that described above. This kaempferol was also
identified in Crocus sativus in a recent work of Goupy et al [40]. In the case of the other two
compounds, their MS spectra showed the deprotonated molecular ions at m/z 771.1811 (at 4.8
min) and 771.1880 (at 5.9 min), characteristic of kaempferols glycosylated with three
hexoses. As described above, in the first case, the presence of the ion with m/z 447.0854 and
the absence of an ion at m/z 429 confirmed that it was a tri-O-glucoside (i.e., kaempferol
3,7,4´-O-triglucoside), while for the second compound the presence of the ion with m/z
429.0736 and the absence of an ion at m/z 447 confirmed that it was the kaempferol 3-O-
sophoroside-7-O-glucoside.
Finally, three standards were analyzed in order to confirm them as markers of
authenticity, geranic acid at 6.4 min, and genistein and baicalein at 11.8 min. The latter two
could not be confirmed for not matching their retention times and MS/MS spectra with the
results obtained in samples of saffron, but the Geranic acid tentatively identified in Table 2
as marker from [C10H16O2+H]+ ion (m/z 169.1196), was confirmed as Geranyl-O-glucoside
after analysis of commercially available geranic acid and studies focused on MS/MS analysis.
When geranic acid was analyzed its retention time was significantly higher (15.8 min) than
that obtained from the sample (6.4 min), but the same MS/MS mass spectra (product ions
spectra) of the precursor ion at m/z 169 were obtained from standard and sample (see Figure
6A, 6B). A detailed study of MS spectra obtained with the saffron sample at 6.4 min (see
Figure 6C) allowed to observe an ion of small intensity at m/z 331.1717corresponding to
glycoside of geranic acid with very good mass accuracy (less 4 mDa), along with the
protonated ion of the geranic acid at m/z 169.1196 and two of its characteristic fragments at
32
m/z 151.1090 and m/z 123.1151. This result justifies the differences between the retention
times, because as it was expected, the more polar glycoside of geranic was eluted earlier than
geranic acid. Therefore, using the available experimental MS and MS/MS data it is possible to
accomplish structural confirmation of Geranyl-O-glucoside as authenticity marker of saffron
according to the fragmentation pattern shown in Figure 7 based on the MS/MS spectra in
Figure 6.
Figure 6. MS/MS spectra in ESI+ of the precursor ion with m/z 169.1 for geranic acid standard eluted at 15.8
min (A) and for a compound of a saffron sample eluted at 6.4 min (B). MS spectra in ESI+ for the same
compound of a saffron sample eluted at 6.4 min tentatively identified as geranic acid in a saffron sample (C).
33
Figure 7. Fragmentation pattern for geranyl-O-glucoside with data related with the molecular
formulas and their theoretical accurate masses.
4. Conclusions
The comprehensive and non-targeted LC–MS metabolomic fingerprinting coupled to
chemometric methods was demonstrated for the first time to be a powerful tool for saffron
authenticity testing. An attempt to identify novel marker compounds was carried out based on
MS/MS mass spectra obtained within the LC-(QTOF)MS analysis. Using a combination of
34
experimental data and information available in scientific literature, and mass spectral
databases, six novel metabolites related to metabolism of kaempferol and geranic acid as
glycosides (kaempferol 3-O-glucoside, kaempferol 3-O-sophoroside, kaempferol 3,7-O-
diglucoside, kaempferol 3,7,4´-O-triglucoside, kaempferol 3-O-sophoroside-7-O-glucoside,
and geranyl-O-glucoside) were identified.
The fact that all possible markers obtained from the comparison between authentic vs.
suspicious samples were regulated in authentic group implies that the suspicious samples
were not adulterated with common elements which may serve as markers of adulteration. In
addition, the large heterogeneity in adulterated samples increased the value of authenticity
markers in saffron. These authenticity markers of saffron have the potential to be a useful tool
for detecting novel adulteration practices, as more advanced and sophisticated adulteration
methods are continuously developing.
However, no marker related to geographical origin was tentatively identified, so a
larger set of samples with differences in origin (environmental conditions such as diverse soil
types, cultivation environments, altitude, etc.) would be required to validate this model and
assign an appropriate confidence level before it could be concluded the provenience of a
saffron sample.
In conclusion, the results obtained in this work demonstrate that metabolomics, in
conjunction with a comprehensive database, has a great potential as a screening tool for the
detection of food fraud, and may be used in the future to enable a rapid reaction in the global
saffron market and to help regulators to stay one step ahead of fraudsters.
Acknowledgements
Authors thank financial support from the Spanish Ministry of Science and Innovation
(project CTQ2009-09022), the Comunidad of Madrid (Spain) and European funding from
35
FEDER program (project S2013/ABI-3028, AVANSECAL-CM), and the University of
Alcalá (project UAH2011/EXP-020). M. Guijarro-Díez thanks the Ministry of Science and
Innovation for his predoctoral contract (BES-2010-033465). Authors also thank the kind gift
of saffron samples by Carmencita Company and the group of Prof. Coral Barbas for the one
month stay of M.G.D. in her laboratory to work with the different chemometric techniques
used in this work.
References
1 Melnyk JP, Wang S, Marcone MF (2010) Chemical and biological properties of the
world’s most expensive spice: Saffron. Food Res Int 43:1981-1989.
2 Winterhalter P, Straubinger MS (2000) Saffron-renewed interest in an ancient spice. Food
Rev Int 16:39-59.
3 Hagh-Nazari S, Keifi N (2007) Saffron and various fraud manners in its production and
trades. Acta Hortic 739:411-416.
4 Kanti R, Sushma K, Ghandi R (2011) Adulteration and marketing of saffron (Crocus
sativus L.). J Trop Med Plants 12:135-140.
5 Carmona M, Zalacain A, Sanchez AM, Novella JL, Alonso GL (2006) Crocetin ester,
picrocrocin and its related compounds present in Crocus sativus stigmas and Gardenia
Jasminoides fruits. Tentative identification of seven new compounds by LC-ESI-MS. J
Agr Food Chem 54:973-979.
6 Sanchez AM, Maggi L, Carmona M, Alonso GL (2011) In: Ebeler S (ed) Progress in
authentication of food and wine. American Chemical Society, Washington DC.
7 International standard ISO 3632-2: Saffron (Crocus sativus L.) test methods.
8 International standard ISO 3632-1: Saffron (Crocus sativus L.) specification.
36
9 Sabatino L, Scordino M, Gargano M, Belligno A, Traulo P, Gagliano G (2011)
HPLC/PDA/ESI-MS evaluation of saffron (Crocus sativus L.) adulteration. Nat Prod
Commun 6:1873-1876.
10 Zalacain A, Ordoudi AS, Blazquez I, Diaz-olaza EM, Carmona M, Tsimldou MZ (2005)
Screening method for the detection of artificial colours in saffron using derivative UV-Vis
spectrometry after precipitation of crocetin. Food Addit Contam 22:607-615.
11 Sanchez AM, Carmona M, Zalacain A, Carot TM, Jabaloyes JM, Alonso GL (2008) Rapid
determination of crocetin esters and picrocrocin from saffron spice (Crocus sativus L.)
using UV visible spectrophotometry for quality control. J Agr Food Chem 56:3167-3175.
12 Maggi L, Sanchez MA, Carmona M, Kanakis DC, Anastasaki E, Tarantilis AP (2011)
Rapid determination of safranal in the quality control of saffron spice (Crocus Sativus L).
Food Chem 127:369-373.
13 Zalacain A, Ordoudi SA, Blazquez I, Diaz-Plaza EM, Carmona M, Blazquez MZI (2005)
Near-infrared spectroscopy in saffron quality control: determination of chemical
composition and geographical origin. J Agr Food Chem 53:9337-9341.
14 Assimiadis MK, Tarantilis PA, Polissiou MG (1998) UV-vis FT-Raman and 1H NMR
spectroscopies of cisetrans carotenoids from saffron (Crocus sativus L). J Appl
Spectroscopy 52:519-522.
15 Tarantilis PA, Polissiou MG (2004) Chemical analysis and antitumor activity of natural
and semi-natural carotenoids of saffron. Acta Hortic 650:447-461.
16 Zougagh M, Simonet BM, Rios A, Valcarcel M (2005) Use of nonaqueous capillary
electrophoresis for the quality control of commercial saffron samples. J Chrom A
1085:293-298.
37
17 Caballero-Ortega H, Pereda-Miranda R, Abdullaex FI (2007) HPLC quantification of
mayor active components from 11 different saffron (Crocus sativus L.) sources. Food
Chem 100:1126-1131.
18 Lage M, Cantrell CL (2009) Quantification of saffron (Crocus sativus L.) metabolites
crocins, picrocrocin and safranal for quality determination of the spice grown under
different environmental Moroccan conditions. Sci Hortic 121:366-373.
19 Loskutov AV, Beninger CW, Hoseld GL, Sink KC (2000) Development of an improved
procedure for extraction and quantitation of safranal in stigmas of Crocus sativus L. using
high performance liquid chromatography. Food Chem 69:87-95.
20 Sabatino L, Scordino M, Gargano M, Belligno A, Traulo P, Gagliano G (2011)
HPLC/PDA/ESI-MS evaluation of saffron (Crocus sativus L.) adulteration. Nat Prod
Commun 6:1873-1876.
21 Javanmardi N, Bagheri A, Moshtaghi N, Sharifi A, Hemati-Kakhki A (2011)
Identification of safflower as a fraud in commercial saffron using RAPD/SCAR marker. J
Cell Mol Res 3:31-37.
22 Marieschi M, Torelli A, Bruni R (2012) Quality control of saffron (Crocus sativus L).
Development of SCAR markers for the detection of plant adulterants used as bulking
agents. J Agr Food Chem 60:10998-11004.
23 Babaei S, Talebi M, Bahar M (2014) Developing an SCAR and ITS reliable multiplex
PCR-based assay for safflower adulterant detection in saffron samples. Food Control
35:323-328.
24 Torelli A, Marieschi M, Bruni R (2014) Authentication of saffron (Crocus sativus L.) in
different processed, retail products by means of SCAR markers. Food Control 36:126-
131.
38
25 Cubero-Leon E, Peñalver R, Maquet A (2014) Review on metabolomics for food
authentication. Food Res Int 60:95-104.
26 Antignac JP, Courant F, Pinel G, Bichon E, Monteau F, Elliott C, Le Bizec B (2011) Mass
spectrometry-based metabolomics applied to the chemical safety of food. Trends Anal
Chem 30:292-301.
27 Hu C, Xu G (2013) Mass-spectrometry-based metabolomics analysis for foodomics.
Trends Anal Chem 52:36-46.
28 Petrakis EA, Cagliani LR, Polissiou MG, Consonni R (2015) Evaluation of saffron
(Crocus sativus L.) adulteration with plant adulterants by 1H NMR metabolite
fingerprinting. Food Chem 173:890-896.
29 Ciborowski M, Teul J, Martin-Ventura JL, Egido J, Barbas C (2012) Metabolomics with
LC–QTOF-MS permits the prediction of disease stage in aortic abdominal aneurysm
based on plasma metabolic fingerprint. PLoS One. Doi: 10.1371/journal.pone.0031982.
30 Westerhuis J, Hoefsloot H, Smit S, Vis D, Smilde A, van Velzen E, van Duijnhoven J,
van Dorsten F (2004) Assessment of PLSDA cross validation. Metabolomics 4:81-89.
31 El Rammouz R, Letisse F, Durand S, Portais JC, Moussa ZW, Fernandez X (2010)
Analysis of skeletal muscle metabolome: evaluation of extraction methods for targeted
metabolite quantification using liquid chromatography tandem mass spectrometry. Anal
Biochem 398:169-177.
32 Bruce S, Tavazzi I, Parisod V, Rezzi S, Kochhar S, Guy P (2009) Investigation of human
blood plasma sample preparation for performing metabolomics using ultrahigh
performance liquid chromatography/mass spectrometry. Anal Chem 81:3285-3296.
33 Berrueta LA, Alfonso-Salces RM, Hérberger K (2007) Supervised pattern recognition in
food analysis. J Chrom A 1158:196-214.
39
34 Trygg J (2002) Orthogonal projections to latent structures (O-PLS). J Chemometr 16:119-
128.
35 Bylesjo M, Rantalainen M, Cloarec O, Nicholson JK, Holmes E, Trygg J (2006) OPLS
discriminant analysis: combining the strengths of PLS‐DA and SIMCA classification. J
Chemometr 20:341-351.
36 Westerhuis JA, Velzen EJJ, Hoefsloot HCJ, Smilde AK (2010) Multivariate paired data
analysis: multilevel PLSDA versus OPLDA. Metabolomics 6:119-128.
37 van den Berg RA, Hoefsloot HC, Westerhuis JA, Smilde AK, van der Werf MJ (2006)
Centering, scaling, and transformations: improving the biological information content of
metabolomics data. BMC Genomics 7:142-147.
38 Rubingh CM, Bijlsma S, Derks E, Bobeldijk I, Verheij ER (2006) Assessing the
performance of statistical validation tools for megavariate metabolomics data.
Metabolomics 2:53-61.
39 Nørbæk R, Brandt K, Nielsen JK, Ørgaard M, Jacobsen N (2002) Flower pigment
composition of Crocus species and cultivars used for a chemotaxonomic investigation.
Biochem Syst Ecol 30:763-791.
40 Goupy P, Vian MA, Chemat F, Caris-Veyrat C (2013) Identification and quantification of
flavonols, anthocyanins and lutein diesters in tepals of Crocus sativus by ultra
performance liquid chromatography coupled to diode array and ion trap mass
spectrometry detections. Ind Crop Prod 44:496-510.
41 Ferreres F, Llorach R, Gil-Izquierdo A (2004) Characterization of the interglycosidic
linkage in di-, tri-, tetra- and pentaglycosylated flavonoids and differentiation of positional
isomers by liquid chromatography/electrospray ionization tandem mass spectrometry. J
Mass Spectrom 39:312-321.
40
42 Serrano-Diaz J, Sanchez AM, Martinez-Tomé M, Winterhalter P, Alonso GL (2014)
Flavonoid Determination in the Quality Control of Floral Bioresidues from Crocus sativus
L. J Agric Food Chem 62:3125-3133.
43 Straubinger M, Jezussek M, Waibel R, Winterhalter P (1997) Novel glycosidic
constituents from saffron. J Agr Food Chem 45:1678-1681.
44 Montoro P, Tuberoso CIG, Maldini M, Cabras P, Pizza C (2008) Qualitative profile and
quantitative determination of flavonoids from Crocus sativus L. Petals by LC–MS/MS.
Nat Prod Commun 3:2013-2016.
45 Termentzi A, Kokkalou E (2008) LC–DAD–MS (ESI+) analysis and antioxidant capacity
of Crocus sativus petal extracts. Planta Med 74:573-581.
46 Vignolini P, Heimler D, Pinelli P, Ieri F, Sciullo A, Romani A (2008) Characterization of
by-products of saffron (Crocus sativus L.) production. Nat Prod Commun 3:1959-1962.
47 Carmona M, Sánchez AM, Ferreres F, Zalacain A, Tomás-Barberán F, Alonso GL (2007)
Identification of the flavonoid fraction in saffron spice by LC/DAD/MS/MS: Comparative
study of samples from different geographical origins. Food Chem 100:445-450.
48 Harborne JB (1999) In: Baxter H (ed) The Handbook of Natural Flavonoids. Wiley, New
York.